Nobody Reads the Code Anymore: Quality Engineering for the Software Factory

We produce more code than ever, and we read less of it than ever. Agents write it, review it, ship it. Everyone is building their own software factory, and all the effort goes into making it produce more. But careful review of code we understand was the quality model for decades, and it doesn’t survive this shift. What replaces review is verification: proving the behavior instead of reading the diff. A factory optimizes toward whatever target you give it. Encode the target sloppily and you get noise. Your business knowledge and product intent have to become outcomes agents can verify, because that’s what they optimize toward. Your test suite is the loss function of your software factory. Drawing on how we build and test at QA.tech, with agents at every step, I’ll show what verification looks like in practice: testing agents navigating real applications, and evals for systems whose output you can’t predict. You’ll learn how testing agents verify real applications, how to encode intent into verifiable outcomes, and how to catch regressions without golden outputs.